include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit...
[safe/jmp/linux-2.6] / drivers / infiniband / core / uverbs_main.c
index ef2312a..fb35262 100644 (file)
@@ -1,6 +1,6 @@
 /*
  * Copyright (c) 2005 Topspin Communications.  All rights reserved.
- * Copyright (c) 2005 Cisco Systems.  All rights reserved.
+ * Copyright (c) 2005, 2006 Cisco Systems.  All rights reserved.
  * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
  * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
  * Copyright (c) 2005 PathScale, Inc. All rights reserved.
@@ -32,8 +32,6 @@
  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
- *
- * $Id: uverbs_main.c 2733 2005-06-28 19:14:34Z roland $
  */
 
 #include <linux/module.h>
 #include <linux/err.h>
 #include <linux/fs.h>
 #include <linux/poll.h>
+#include <linux/sched.h>
 #include <linux/file.h>
-#include <linux/mount.h>
+#include <linux/cdev.h>
+#include <linux/anon_inodes.h>
+#include <linux/slab.h>
 
 #include <asm/uaccess.h>
 
@@ -53,8 +54,6 @@ MODULE_AUTHOR("Roland Dreier");
 MODULE_DESCRIPTION("InfiniBand userspace verbs access");
 MODULE_LICENSE("Dual BSD/GPL");
 
-#define INFINIBANDEVENTFS_MAGIC        0x49426576      /* "IBev" */
-
 enum {
        IB_UVERBS_MAJOR       = 231,
        IB_UVERBS_BASE_MINOR  = 192,
@@ -63,7 +62,9 @@ enum {
 
 #define IB_UVERBS_BASE_DEV     MKDEV(IB_UVERBS_MAJOR, IB_UVERBS_BASE_MINOR)
 
-DECLARE_MUTEX(ib_uverbs_idr_mutex);
+static struct class *uverbs_class;
+
+DEFINE_SPINLOCK(ib_uverbs_idr_lock);
 DEFINE_IDR(ib_uverbs_pd_idr);
 DEFINE_IDR(ib_uverbs_mr_idr);
 DEFINE_IDR(ib_uverbs_mw_idr);
@@ -72,112 +73,182 @@ DEFINE_IDR(ib_uverbs_cq_idr);
 DEFINE_IDR(ib_uverbs_qp_idr);
 DEFINE_IDR(ib_uverbs_srq_idr);
 
-static spinlock_t map_lock;
+static DEFINE_SPINLOCK(map_lock);
 static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
 
 static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
                                     const char __user *buf, int in_len,
                                     int out_len) = {
-       [IB_USER_VERBS_CMD_GET_CONTEXT]         = ib_uverbs_get_context,
-       [IB_USER_VERBS_CMD_QUERY_DEVICE]        = ib_uverbs_query_device,
-       [IB_USER_VERBS_CMD_QUERY_PORT]          = ib_uverbs_query_port,
-       [IB_USER_VERBS_CMD_ALLOC_PD]            = ib_uverbs_alloc_pd,
-       [IB_USER_VERBS_CMD_DEALLOC_PD]          = ib_uverbs_dealloc_pd,
-       [IB_USER_VERBS_CMD_REG_MR]              = ib_uverbs_reg_mr,
-       [IB_USER_VERBS_CMD_DEREG_MR]            = ib_uverbs_dereg_mr,
+       [IB_USER_VERBS_CMD_GET_CONTEXT]         = ib_uverbs_get_context,
+       [IB_USER_VERBS_CMD_QUERY_DEVICE]        = ib_uverbs_query_device,
+       [IB_USER_VERBS_CMD_QUERY_PORT]          = ib_uverbs_query_port,
+       [IB_USER_VERBS_CMD_ALLOC_PD]            = ib_uverbs_alloc_pd,
+       [IB_USER_VERBS_CMD_DEALLOC_PD]          = ib_uverbs_dealloc_pd,
+       [IB_USER_VERBS_CMD_REG_MR]              = ib_uverbs_reg_mr,
+       [IB_USER_VERBS_CMD_DEREG_MR]            = ib_uverbs_dereg_mr,
        [IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL] = ib_uverbs_create_comp_channel,
-       [IB_USER_VERBS_CMD_CREATE_CQ]           = ib_uverbs_create_cq,
-       [IB_USER_VERBS_CMD_POLL_CQ]             = ib_uverbs_poll_cq,
-       [IB_USER_VERBS_CMD_REQ_NOTIFY_CQ]       = ib_uverbs_req_notify_cq,
-       [IB_USER_VERBS_CMD_DESTROY_CQ]          = ib_uverbs_destroy_cq,
-       [IB_USER_VERBS_CMD_CREATE_QP]           = ib_uverbs_create_qp,
-       [IB_USER_VERBS_CMD_MODIFY_QP]           = ib_uverbs_modify_qp,
-       [IB_USER_VERBS_CMD_DESTROY_QP]          = ib_uverbs_destroy_qp,
-       [IB_USER_VERBS_CMD_POST_SEND]           = ib_uverbs_post_send,
-       [IB_USER_VERBS_CMD_POST_RECV]           = ib_uverbs_post_recv,
-       [IB_USER_VERBS_CMD_POST_SRQ_RECV]       = ib_uverbs_post_srq_recv,
-       [IB_USER_VERBS_CMD_CREATE_AH]           = ib_uverbs_create_ah,
-       [IB_USER_VERBS_CMD_DESTROY_AH]          = ib_uverbs_destroy_ah,
-       [IB_USER_VERBS_CMD_ATTACH_MCAST]        = ib_uverbs_attach_mcast,
-       [IB_USER_VERBS_CMD_DETACH_MCAST]        = ib_uverbs_detach_mcast,
-       [IB_USER_VERBS_CMD_CREATE_SRQ]          = ib_uverbs_create_srq,
-       [IB_USER_VERBS_CMD_MODIFY_SRQ]          = ib_uverbs_modify_srq,
-       [IB_USER_VERBS_CMD_DESTROY_SRQ]         = ib_uverbs_destroy_srq,
+       [IB_USER_VERBS_CMD_CREATE_CQ]           = ib_uverbs_create_cq,
+       [IB_USER_VERBS_CMD_RESIZE_CQ]           = ib_uverbs_resize_cq,
+       [IB_USER_VERBS_CMD_POLL_CQ]             = ib_uverbs_poll_cq,
+       [IB_USER_VERBS_CMD_REQ_NOTIFY_CQ]       = ib_uverbs_req_notify_cq,
+       [IB_USER_VERBS_CMD_DESTROY_CQ]          = ib_uverbs_destroy_cq,
+       [IB_USER_VERBS_CMD_CREATE_QP]           = ib_uverbs_create_qp,
+       [IB_USER_VERBS_CMD_QUERY_QP]            = ib_uverbs_query_qp,
+       [IB_USER_VERBS_CMD_MODIFY_QP]           = ib_uverbs_modify_qp,
+       [IB_USER_VERBS_CMD_DESTROY_QP]          = ib_uverbs_destroy_qp,
+       [IB_USER_VERBS_CMD_POST_SEND]           = ib_uverbs_post_send,
+       [IB_USER_VERBS_CMD_POST_RECV]           = ib_uverbs_post_recv,
+       [IB_USER_VERBS_CMD_POST_SRQ_RECV]       = ib_uverbs_post_srq_recv,
+       [IB_USER_VERBS_CMD_CREATE_AH]           = ib_uverbs_create_ah,
+       [IB_USER_VERBS_CMD_DESTROY_AH]          = ib_uverbs_destroy_ah,
+       [IB_USER_VERBS_CMD_ATTACH_MCAST]        = ib_uverbs_attach_mcast,
+       [IB_USER_VERBS_CMD_DETACH_MCAST]        = ib_uverbs_detach_mcast,
+       [IB_USER_VERBS_CMD_CREATE_SRQ]          = ib_uverbs_create_srq,
+       [IB_USER_VERBS_CMD_MODIFY_SRQ]          = ib_uverbs_modify_srq,
+       [IB_USER_VERBS_CMD_QUERY_SRQ]           = ib_uverbs_query_srq,
+       [IB_USER_VERBS_CMD_DESTROY_SRQ]         = ib_uverbs_destroy_srq,
 };
 
-static struct vfsmount *uverbs_event_mnt;
-
 static void ib_uverbs_add_one(struct ib_device *device);
 static void ib_uverbs_remove_one(struct ib_device *device);
 
-static int ib_dealloc_ucontext(struct ib_ucontext *context)
+static void ib_uverbs_release_dev(struct kref *ref)
+{
+       struct ib_uverbs_device *dev =
+               container_of(ref, struct ib_uverbs_device, ref);
+
+       complete(&dev->comp);
+}
+
+static void ib_uverbs_release_event_file(struct kref *ref)
+{
+       struct ib_uverbs_event_file *file =
+               container_of(ref, struct ib_uverbs_event_file, ref);
+
+       kfree(file);
+}
+
+void ib_uverbs_release_ucq(struct ib_uverbs_file *file,
+                         struct ib_uverbs_event_file *ev_file,
+                         struct ib_ucq_object *uobj)
+{
+       struct ib_uverbs_event *evt, *tmp;
+
+       if (ev_file) {
+               spin_lock_irq(&ev_file->lock);
+               list_for_each_entry_safe(evt, tmp, &uobj->comp_list, obj_list) {
+                       list_del(&evt->list);
+                       kfree(evt);
+               }
+               spin_unlock_irq(&ev_file->lock);
+
+               kref_put(&ev_file->ref, ib_uverbs_release_event_file);
+       }
+
+       spin_lock_irq(&file->async_file->lock);
+       list_for_each_entry_safe(evt, tmp, &uobj->async_list, obj_list) {
+               list_del(&evt->list);
+               kfree(evt);
+       }
+       spin_unlock_irq(&file->async_file->lock);
+}
+
+void ib_uverbs_release_uevent(struct ib_uverbs_file *file,
+                             struct ib_uevent_object *uobj)
+{
+       struct ib_uverbs_event *evt, *tmp;
+
+       spin_lock_irq(&file->async_file->lock);
+       list_for_each_entry_safe(evt, tmp, &uobj->event_list, obj_list) {
+               list_del(&evt->list);
+               kfree(evt);
+       }
+       spin_unlock_irq(&file->async_file->lock);
+}
+
+static void ib_uverbs_detach_umcast(struct ib_qp *qp,
+                                   struct ib_uqp_object *uobj)
+{
+       struct ib_uverbs_mcast_entry *mcast, *tmp;
+
+       list_for_each_entry_safe(mcast, tmp, &uobj->mcast_list, list) {
+               ib_detach_mcast(qp, &mcast->gid, mcast->lid);
+               list_del(&mcast->list);
+               kfree(mcast);
+       }
+}
+
+static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
+                                     struct ib_ucontext *context)
 {
        struct ib_uobject *uobj, *tmp;
 
        if (!context)
                return 0;
 
-       down(&ib_uverbs_idr_mutex);
+       context->closing = 1;
 
        list_for_each_entry_safe(uobj, tmp, &context->ah_list, list) {
-               struct ib_ah *ah = idr_find(&ib_uverbs_ah_idr, uobj->id);
-               idr_remove(&ib_uverbs_ah_idr, uobj->id);
+               struct ib_ah *ah = uobj->object;
+
+               idr_remove_uobj(&ib_uverbs_ah_idr, uobj);
                ib_destroy_ah(ah);
-               list_del(&uobj->list);
                kfree(uobj);
        }
 
        list_for_each_entry_safe(uobj, tmp, &context->qp_list, list) {
-               struct ib_qp *qp = idr_find(&ib_uverbs_qp_idr, uobj->id);
-               idr_remove(&ib_uverbs_qp_idr, uobj->id);
+               struct ib_qp *qp = uobj->object;
+               struct ib_uqp_object *uqp =
+                       container_of(uobj, struct ib_uqp_object, uevent.uobject);
+
+               idr_remove_uobj(&ib_uverbs_qp_idr, uobj);
+               ib_uverbs_detach_umcast(qp, uqp);
                ib_destroy_qp(qp);
-               list_del(&uobj->list);
-               kfree(container_of(uobj, struct ib_uevent_object, uobject));
+               ib_uverbs_release_uevent(file, &uqp->uevent);
+               kfree(uqp);
        }
 
        list_for_each_entry_safe(uobj, tmp, &context->cq_list, list) {
-               struct ib_cq *cq = idr_find(&ib_uverbs_cq_idr, uobj->id);
-               idr_remove(&ib_uverbs_cq_idr, uobj->id);
+               struct ib_cq *cq = uobj->object;
+               struct ib_uverbs_event_file *ev_file = cq->cq_context;
+               struct ib_ucq_object *ucq =
+                       container_of(uobj, struct ib_ucq_object, uobject);
+
+               idr_remove_uobj(&ib_uverbs_cq_idr, uobj);
                ib_destroy_cq(cq);
-               list_del(&uobj->list);
-               kfree(container_of(uobj, struct ib_ucq_object, uobject));
+               ib_uverbs_release_ucq(file, ev_file, ucq);
+               kfree(ucq);
        }
 
        list_for_each_entry_safe(uobj, tmp, &context->srq_list, list) {
-               struct ib_srq *srq = idr_find(&ib_uverbs_srq_idr, uobj->id);
-               idr_remove(&ib_uverbs_srq_idr, uobj->id);
+               struct ib_srq *srq = uobj->object;
+               struct ib_uevent_object *uevent =
+                       container_of(uobj, struct ib_uevent_object, uobject);
+
+               idr_remove_uobj(&ib_uverbs_srq_idr, uobj);
                ib_destroy_srq(srq);
-               list_del(&uobj->list);
-               kfree(container_of(uobj, struct ib_uevent_object, uobject));
+               ib_uverbs_release_uevent(file, uevent);
+               kfree(uevent);
        }
 
        /* XXX Free MWs */
 
        list_for_each_entry_safe(uobj, tmp, &context->mr_list, list) {
-               struct ib_mr *mr = idr_find(&ib_uverbs_mr_idr, uobj->id);
-               struct ib_device *mrdev = mr->device;
-               struct ib_umem_object *memobj;
+               struct ib_mr *mr = uobj->object;
 
-               idr_remove(&ib_uverbs_mr_idr, uobj->id);
+               idr_remove_uobj(&ib_uverbs_mr_idr, uobj);
                ib_dereg_mr(mr);
-
-               memobj = container_of(uobj, struct ib_umem_object, uobject);
-               ib_umem_release_on_close(mrdev, &memobj->umem);
-
-               list_del(&uobj->list);
-               kfree(memobj);
+               kfree(uobj);
        }
 
        list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
-               struct ib_pd *pd = idr_find(&ib_uverbs_pd_idr, uobj->id);
-               idr_remove(&ib_uverbs_pd_idr, uobj->id);
+               struct ib_pd *pd = uobj->object;
+
+               idr_remove_uobj(&ib_uverbs_pd_idr, uobj);
                ib_dealloc_pd(pd);
-               list_del(&uobj->list);
                kfree(uobj);
        }
 
-       up(&ib_uverbs_idr_mutex);
-
        return context->device->dealloc_ucontext(context);
 }
 
@@ -187,6 +258,8 @@ static void ib_uverbs_release_file(struct kref *ref)
                container_of(ref, struct ib_uverbs_file, ref);
 
        module_put(file->device->ib_dev->owner);
+       kref_put(&file->device->ref, ib_uverbs_release_dev);
+
        kfree(file);
 }
 
@@ -261,14 +334,6 @@ static unsigned int ib_uverbs_event_poll(struct file *filp,
        return pollflags;
 }
 
-void ib_uverbs_release_event_file(struct kref *ref)
-{
-       struct ib_uverbs_event_file *file =
-               container_of(ref, struct ib_uverbs_event_file, ref);
-
-       kfree(file);
-}
-
 static int ib_uverbs_event_fasync(int fd, struct file *filp, int on)
 {
        struct ib_uverbs_event_file *file = filp->private_data;
@@ -282,7 +347,7 @@ static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
        struct ib_uverbs_event *entry, *tmp;
 
        spin_lock_irq(&file->lock);
-       file->file = NULL;
+       file->is_closed = 1;
        list_for_each_entry_safe(entry, tmp, &file->event_list, list) {
                if (entry->counter)
                        list_del(&entry->obj_list);
@@ -290,8 +355,6 @@ static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
        }
        spin_unlock_irq(&file->lock);
 
-       ib_uverbs_event_fasync(-1, filp, 0);
-
        if (file->is_async) {
                ib_unregister_event_handler(&file->uverbs_file->event_handler);
                kref_put(&file->uverbs_file->ref, ib_uverbs_release_file);
@@ -301,9 +364,9 @@ static int ib_uverbs_event_close(struct inode *inode, struct file *filp)
        return 0;
 }
 
-static struct file_operations uverbs_event_fops = {
+static const struct file_operations uverbs_event_fops = {
        .owner   = THIS_MODULE,
-       .read    = ib_uverbs_event_read,
+       .read    = ib_uverbs_event_read,
        .poll    = ib_uverbs_event_poll,
        .release = ib_uverbs_event_close,
        .fasync  = ib_uverbs_event_fasync
@@ -320,7 +383,7 @@ void ib_uverbs_comp_handler(struct ib_cq *cq, void *cq_context)
                return;
 
        spin_lock_irqsave(&file->lock, flags);
-       if (!file->file) {
+       if (file->is_closed) {
                spin_unlock_irqrestore(&file->lock, flags);
                return;
        }
@@ -353,7 +416,7 @@ static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
        unsigned long flags;
 
        spin_lock_irqsave(&file->async_file->lock, flags);
-       if (!file->async_file->file) {
+       if (file->async_file->is_closed) {
                spin_unlock_irqrestore(&file->async_file->lock, flags);
                return;
        }
@@ -379,16 +442,12 @@ static void ib_uverbs_async_handler(struct ib_uverbs_file *file,
 
 void ib_uverbs_cq_event_handler(struct ib_event *event, void *context_ptr)
 {
-       struct ib_uverbs_event_file *ev_file = context_ptr;
-       struct ib_ucq_object *uobj;
+       struct ib_ucq_object *uobj = container_of(event->element.cq->uobject,
+                                                 struct ib_ucq_object, uobject);
 
-       uobj = container_of(event->element.cq->uobject,
-                           struct ib_ucq_object, uobject);
-
-       ib_uverbs_async_handler(ev_file->uverbs_file, uobj->uobject.user_handle,
+       ib_uverbs_async_handler(uobj->uverbs_file, uobj->uobject.user_handle,
                                event->event, &uobj->async_list,
                                &uobj->async_events_reported);
-                               
 }
 
 void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr)
@@ -426,11 +485,10 @@ void ib_uverbs_event_handler(struct ib_event_handler *handler,
 }
 
 struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
-                                       int is_async, int *fd)
+                                       int is_async)
 {
        struct ib_uverbs_event_file *ev_file;
        struct file *filp;
-       int ret;
 
        ev_file = kmalloc(sizeof *ev_file, GFP_KERNEL);
        if (!ev_file)
@@ -443,42 +501,14 @@ struct file *ib_uverbs_alloc_event_file(struct ib_uverbs_file *uverbs_file,
        ev_file->uverbs_file = uverbs_file;
        ev_file->async_queue = NULL;
        ev_file->is_async    = is_async;
+       ev_file->is_closed   = 0;
 
-       *fd = get_unused_fd();
-       if (*fd < 0) {
-               ret = *fd;
-               goto err;
-       }
-
-       filp = get_empty_filp();
-       if (!filp) {
-               ret = -ENFILE;
-               goto err_fd;
-       }
-
-       ev_file->file      = filp;
-
-       /*
-        * fops_get() can't fail here, because we're coming from a
-        * system call on a uverbs file, which will already have a
-        * module reference.
-        */
-       filp->f_op         = fops_get(&uverbs_event_fops);
-       filp->f_vfsmnt     = mntget(uverbs_event_mnt);
-       filp->f_dentry     = dget(uverbs_event_mnt->mnt_root);
-       filp->f_mapping    = filp->f_dentry->d_inode->i_mapping;
-       filp->f_flags      = O_RDONLY;
-       filp->f_mode       = FMODE_READ;
-       filp->private_data = ev_file;
+       filp = anon_inode_getfile("[infinibandevent]", &uverbs_event_fops,
+                                 ev_file, O_RDONLY);
+       if (IS_ERR(filp))
+               kfree(ev_file);
 
        return filp;
-
-err_fd:
-       put_unused_fd(*fd);
-
-err:
-       kfree(ev_file);
-       return ERR_PTR(ret);
 }
 
 /*
@@ -528,14 +558,16 @@ static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
 
        if (hdr.command < 0                             ||
            hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
-           !uverbs_cmd_table[hdr.command]              ||
-           !(file->device->ib_dev->uverbs_cmd_mask & (1ull << hdr.command)))
+           !uverbs_cmd_table[hdr.command])
                return -EINVAL;
 
        if (!file->ucontext &&
            hdr.command != IB_USER_VERBS_CMD_GET_CONTEXT)
                return -EINVAL;
 
+       if (!(file->device->ib_dev->uverbs_cmd_mask & (1ull << hdr.command)))
+               return -ENOSYS;
+
        return uverbs_cmd_table[hdr.command](file, buf + sizeof hdr,
                                             hdr.in_words * 4, hdr.out_words * 4);
 }
@@ -550,55 +582,83 @@ static int ib_uverbs_mmap(struct file *filp, struct vm_area_struct *vma)
                return file->device->ib_dev->mmap(file->ucontext, vma);
 }
 
+/*
+ * ib_uverbs_open() does not need the BKL:
+ *
+ *  - the ib_uverbs_device structures are properly reference counted and
+ *    everything else is purely local to the file being created, so
+ *    races against other open calls are not a problem;
+ *  - there is no ioctl method to race against;
+ *  - the open method will either immediately run -ENXIO, or all
+ *    required initialization will be done.
+ */
 static int ib_uverbs_open(struct inode *inode, struct file *filp)
 {
-       struct ib_uverbs_device *dev =
-               container_of(inode->i_cdev, struct ib_uverbs_device, dev);
+       struct ib_uverbs_device *dev;
        struct ib_uverbs_file *file;
+       int ret;
 
-       if (!try_module_get(dev->ib_dev->owner))
-               return -ENODEV;
+       dev = container_of(inode->i_cdev, struct ib_uverbs_device, cdev);
+       if (dev)
+               kref_get(&dev->ref);
+       else
+               return -ENXIO;
+
+       if (!try_module_get(dev->ib_dev->owner)) {
+               ret = -ENODEV;
+               goto err;
+       }
 
        file = kmalloc(sizeof *file, GFP_KERNEL);
        if (!file) {
-               module_put(dev->ib_dev->owner);
-               return -ENOMEM;
+               ret = -ENOMEM;
+               goto err_module;
        }
 
-       file->device   = dev;
-       file->ucontext = NULL;
+       file->device     = dev;
+       file->ucontext   = NULL;
+       file->async_file = NULL;
        kref_init(&file->ref);
-       init_MUTEX(&file->mutex);
+       mutex_init(&file->mutex);
 
        filp->private_data = file;
 
        return 0;
+
+err_module:
+       module_put(dev->ib_dev->owner);
+
+err:
+       kref_put(&dev->ref, ib_uverbs_release_dev);
+       return ret;
 }
 
 static int ib_uverbs_close(struct inode *inode, struct file *filp)
 {
        struct ib_uverbs_file *file = filp->private_data;
 
-       ib_dealloc_ucontext(file->ucontext);
+       ib_uverbs_cleanup_ucontext(file, file->ucontext);
+
+       if (file->async_file)
+               kref_put(&file->async_file->ref, ib_uverbs_release_event_file);
 
-       kref_put(&file->async_file->ref, ib_uverbs_release_event_file);
        kref_put(&file->ref, ib_uverbs_release_file);
 
        return 0;
 }
 
-static struct file_operations uverbs_fops = {
-       .owner   = THIS_MODULE,
-       .write   = ib_uverbs_write,
-       .open    = ib_uverbs_open,
+static const struct file_operations uverbs_fops = {
+       .owner   = THIS_MODULE,
+       .write   = ib_uverbs_write,
+       .open    = ib_uverbs_open,
        .release = ib_uverbs_close
 };
 
-static struct file_operations uverbs_mmap_fops = {
-       .owner   = THIS_MODULE,
-       .write   = ib_uverbs_write,
+static const struct file_operations uverbs_mmap_fops = {
+       .owner   = THIS_MODULE,
+       .write   = ib_uverbs_write,
        .mmap    = ib_uverbs_mmap,
-       .open    = ib_uverbs_open,
+       .open    = ib_uverbs_open,
        .release = ib_uverbs_close
 };
 
@@ -608,89 +668,115 @@ static struct ib_client uverbs_client = {
        .remove = ib_uverbs_remove_one
 };
 
-static ssize_t show_ibdev(struct class_device *class_dev, char *buf)
+static ssize_t show_ibdev(struct device *device, struct device_attribute *attr,
+                         char *buf)
 {
-       struct ib_uverbs_device *dev =
-               container_of(class_dev, struct ib_uverbs_device, class_dev);
+       struct ib_uverbs_device *dev = dev_get_drvdata(device);
+
+       if (!dev)
+               return -ENODEV;
 
        return sprintf(buf, "%s\n", dev->ib_dev->name);
 }
-static CLASS_DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
+static DEVICE_ATTR(ibdev, S_IRUGO, show_ibdev, NULL);
 
-static ssize_t show_dev_abi_version(struct class_device *class_dev, char *buf)
+static ssize_t show_dev_abi_version(struct device *device,
+                                   struct device_attribute *attr, char *buf)
 {
-       struct ib_uverbs_device *dev =
-               container_of(class_dev, struct ib_uverbs_device, class_dev);
+       struct ib_uverbs_device *dev = dev_get_drvdata(device);
+
+       if (!dev)
+               return -ENODEV;
 
        return sprintf(buf, "%d\n", dev->ib_dev->uverbs_abi_ver);
 }
-static CLASS_DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
+static DEVICE_ATTR(abi_version, S_IRUGO, show_dev_abi_version, NULL);
+
+static CLASS_ATTR_STRING(abi_version, S_IRUGO,
+                        __stringify(IB_USER_VERBS_ABI_VERSION));
 
-static void ib_uverbs_release_class_dev(struct class_device *class_dev)
+static dev_t overflow_maj;
+static DECLARE_BITMAP(overflow_map, IB_UVERBS_MAX_DEVICES);
+
+/*
+ * If we have more than IB_UVERBS_MAX_DEVICES, dynamically overflow by
+ * requesting a new major number and doubling the number of max devices we
+ * support. It's stupid, but simple.
+ */
+static int find_overflow_devnum(void)
 {
-       struct ib_uverbs_device *dev =
-               container_of(class_dev, struct ib_uverbs_device, class_dev);
+       int ret;
 
-       cdev_del(&dev->dev);
-       clear_bit(dev->devnum, dev_map);
-       kfree(dev);
-}
+       if (!overflow_maj) {
+               ret = alloc_chrdev_region(&overflow_maj, 0, IB_UVERBS_MAX_DEVICES,
+                                         "infiniband_verbs");
+               if (ret) {
+                       printk(KERN_ERR "user_verbs: couldn't register dynamic device number\n");
+                       return ret;
+               }
+       }
 
-static struct class uverbs_class = {
-       .name    = "infiniband_verbs",
-       .release = ib_uverbs_release_class_dev
-};
+       ret = find_first_zero_bit(overflow_map, IB_UVERBS_MAX_DEVICES);
+       if (ret >= IB_UVERBS_MAX_DEVICES)
+               return -1;
 
-static ssize_t show_abi_version(struct class *class, char *buf)
-{
-       return sprintf(buf, "%d\n", IB_USER_VERBS_ABI_VERSION);
+       return ret;
 }
-static CLASS_ATTR(abi_version, S_IRUGO, show_abi_version, NULL);
 
 static void ib_uverbs_add_one(struct ib_device *device)
 {
+       int devnum;
+       dev_t base;
        struct ib_uverbs_device *uverbs_dev;
 
        if (!device->alloc_ucontext)
                return;
 
-       uverbs_dev = kmalloc(sizeof *uverbs_dev, GFP_KERNEL);
+       uverbs_dev = kzalloc(sizeof *uverbs_dev, GFP_KERNEL);
        if (!uverbs_dev)
                return;
 
-       memset(uverbs_dev, 0, sizeof *uverbs_dev);
+       kref_init(&uverbs_dev->ref);
+       init_completion(&uverbs_dev->comp);
 
        spin_lock(&map_lock);
-       uverbs_dev->devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
-       if (uverbs_dev->devnum >= IB_UVERBS_MAX_DEVICES) {
+       devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);
+       if (devnum >= IB_UVERBS_MAX_DEVICES) {
                spin_unlock(&map_lock);
-               goto err;
+               devnum = find_overflow_devnum();
+               if (devnum < 0)
+                       goto err;
+
+               spin_lock(&map_lock);
+               uverbs_dev->devnum = devnum + IB_UVERBS_MAX_DEVICES;
+               base = devnum + overflow_maj;
+               set_bit(devnum, overflow_map);
+       } else {
+               uverbs_dev->devnum = devnum;
+               base = devnum + IB_UVERBS_BASE_DEV;
+               set_bit(devnum, dev_map);
        }
-       set_bit(uverbs_dev->devnum, dev_map);
        spin_unlock(&map_lock);
 
        uverbs_dev->ib_dev           = device;
-       uverbs_dev->num_comp_vectors = 1;
+       uverbs_dev->num_comp_vectors = device->num_comp_vectors;
 
-       if (device->mmap)
-               cdev_init(&uverbs_dev->dev, &uverbs_mmap_fops);
-       else
-               cdev_init(&uverbs_dev->dev, &uverbs_fops);
-       uverbs_dev->dev.owner = THIS_MODULE;
-       kobject_set_name(&uverbs_dev->dev.kobj, "uverbs%d", uverbs_dev->devnum);
-       if (cdev_add(&uverbs_dev->dev, IB_UVERBS_BASE_DEV + uverbs_dev->devnum, 1))
-               goto err;
+       cdev_init(&uverbs_dev->cdev, NULL);
+       uverbs_dev->cdev.owner = THIS_MODULE;
+       uverbs_dev->cdev.ops = device->mmap ? &uverbs_mmap_fops : &uverbs_fops;
+       kobject_set_name(&uverbs_dev->cdev.kobj, "uverbs%d", uverbs_dev->devnum);
+       if (cdev_add(&uverbs_dev->cdev, base, 1))
+               goto err_cdev;
 
-       uverbs_dev->class_dev.class = &uverbs_class;
-       uverbs_dev->class_dev.dev   = device->dma_device;
-       uverbs_dev->class_dev.devt  = uverbs_dev->dev.dev;
-       snprintf(uverbs_dev->class_dev.class_id, BUS_ID_SIZE, "uverbs%d", uverbs_dev->devnum);
-       if (class_device_register(&uverbs_dev->class_dev))
+       uverbs_dev->dev = device_create(uverbs_class, device->dma_device,
+                                       uverbs_dev->cdev.dev, uverbs_dev,
+                                       "uverbs%d", uverbs_dev->devnum);
+       if (IS_ERR(uverbs_dev->dev))
                goto err_cdev;
 
-       if (class_device_create_file(&uverbs_dev->class_dev, &class_device_attr_ibdev))
+       if (device_create_file(uverbs_dev->dev, &dev_attr_ibdev))
                goto err_class;
-       if (class_device_create_file(&uverbs_dev->class_dev, &class_device_attr_abi_version))
+       if (device_create_file(uverbs_dev->dev, &dev_attr_abi_version))
                goto err_class;
 
        ib_set_client_data(device, &uverbs_client, uverbs_dev);
@@ -698,13 +784,18 @@ static void ib_uverbs_add_one(struct ib_device *device)
        return;
 
 err_class:
-       class_device_unregister(&uverbs_dev->class_dev);
+       device_destroy(uverbs_class, uverbs_dev->cdev.dev);
 
 err_cdev:
-       cdev_del(&uverbs_dev->dev);
-       clear_bit(uverbs_dev->devnum, dev_map);
+       cdev_del(&uverbs_dev->cdev);
+       if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES)
+               clear_bit(devnum, dev_map);
+       else
+               clear_bit(devnum, overflow_map);
 
 err:
+       kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
+       wait_for_completion(&uverbs_dev->comp);
        kfree(uverbs_dev);
        return;
 }
@@ -716,29 +807,24 @@ static void ib_uverbs_remove_one(struct ib_device *device)
        if (!uverbs_dev)
                return;
 
-       class_device_unregister(&uverbs_dev->class_dev);
-}
+       dev_set_drvdata(uverbs_dev->dev, NULL);
+       device_destroy(uverbs_class, uverbs_dev->cdev.dev);
+       cdev_del(&uverbs_dev->cdev);
 
-static struct super_block *uverbs_event_get_sb(struct file_system_type *fs_type, int flags,
-                                              const char *dev_name, void *data)
-{
-       return get_sb_pseudo(fs_type, "infinibandevent:", NULL,
-                            INFINIBANDEVENTFS_MAGIC);
-}
+       if (uverbs_dev->devnum < IB_UVERBS_MAX_DEVICES)
+               clear_bit(uverbs_dev->devnum, dev_map);
+       else
+               clear_bit(uverbs_dev->devnum - IB_UVERBS_MAX_DEVICES, overflow_map);
 
-static struct file_system_type uverbs_event_fs = {
-       /* No owner field so module can be unloaded */
-       .name    = "infinibandeventfs",
-       .get_sb  = uverbs_event_get_sb,
-       .kill_sb = kill_litter_super
-};
+       kref_put(&uverbs_dev->ref, ib_uverbs_release_dev);
+       wait_for_completion(&uverbs_dev->comp);
+       kfree(uverbs_dev);
+}
 
 static int __init ib_uverbs_init(void)
 {
        int ret;
 
-       spin_lock_init(&map_lock);
-
        ret = register_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES,
                                     "infiniband_verbs");
        if (ret) {
@@ -746,47 +832,29 @@ static int __init ib_uverbs_init(void)
                goto out;
        }
 
-       ret = class_register(&uverbs_class);
-       if (ret) {
+       uverbs_class = class_create(THIS_MODULE, "infiniband_verbs");
+       if (IS_ERR(uverbs_class)) {
+               ret = PTR_ERR(uverbs_class);
                printk(KERN_ERR "user_verbs: couldn't create class infiniband_verbs\n");
                goto out_chrdev;
        }
 
-       ret = class_create_file(&uverbs_class, &class_attr_abi_version);
+       ret = class_create_file(uverbs_class, &class_attr_abi_version.attr);
        if (ret) {
                printk(KERN_ERR "user_verbs: couldn't create abi_version attribute\n");
                goto out_class;
        }
 
-       ret = register_filesystem(&uverbs_event_fs);
-       if (ret) {
-               printk(KERN_ERR "user_verbs: couldn't register infinibandeventfs\n");
-               goto out_class;
-       }
-
-       uverbs_event_mnt = kern_mount(&uverbs_event_fs);
-       if (IS_ERR(uverbs_event_mnt)) {
-               ret = PTR_ERR(uverbs_event_mnt);
-               printk(KERN_ERR "user_verbs: couldn't mount infinibandeventfs\n");
-               goto out_fs;
-       }
-
        ret = ib_register_client(&uverbs_client);
        if (ret) {
                printk(KERN_ERR "user_verbs: couldn't register client\n");
-               goto out_mnt;
+               goto out_class;
        }
 
        return 0;
 
-out_mnt:
-       mntput(uverbs_event_mnt);
-
-out_fs:
-       unregister_filesystem(&uverbs_event_fs);
-
 out_class:
-       class_unregister(&uverbs_class);
+       class_destroy(uverbs_class);
 
 out_chrdev:
        unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
@@ -798,10 +866,17 @@ out:
 static void __exit ib_uverbs_cleanup(void)
 {
        ib_unregister_client(&uverbs_client);
-       mntput(uverbs_event_mnt);
-       unregister_filesystem(&uverbs_event_fs);
-       class_unregister(&uverbs_class);
+       class_destroy(uverbs_class);
        unregister_chrdev_region(IB_UVERBS_BASE_DEV, IB_UVERBS_MAX_DEVICES);
+       if (overflow_maj)
+               unregister_chrdev_region(overflow_maj, IB_UVERBS_MAX_DEVICES);
+       idr_destroy(&ib_uverbs_pd_idr);
+       idr_destroy(&ib_uverbs_mr_idr);
+       idr_destroy(&ib_uverbs_mw_idr);
+       idr_destroy(&ib_uverbs_ah_idr);
+       idr_destroy(&ib_uverbs_cq_idr);
+       idr_destroy(&ib_uverbs_qp_idr);
+       idr_destroy(&ib_uverbs_srq_idr);
 }
 
 module_init(ib_uverbs_init);